The White House has finalized a voluntary framework for reviewing advanced AI models before release, but the testing criteria are staying private. That turns frontier model launches into a new kind of prerelease security review, with open-weight models and outside researchers still sitting largely outside the process.
Booz Allen and OpenAI are partnering to deploy frontier AI for defense, intelligence, critical infrastructure, and commercial operations. The deal shows how OpenAI’s government push is moving from model access and pilots toward implementation through mission contractors.